Parallel Optimization Of Reverse Time Migration For VTI Media

Reverse time migration algorithm plays an important role in energy exploration industry.Its main function is to image complex geological structures through detected seismic wave data and provide convenience for searching for underground energy sources.At present,the reverse time migration algorithm still faces many difficulties in the field of practical application,mainly including: excessive time consumption due to multiple iterations of the algorithm itself,limited experiment scale due to lack of single node memory,and low read and write efficiency due to serial read and write mode.It is of great significance and value to solve these problems and make reverse time migration algorithm face to industrial application.This paper studies the parallel optimization of reverse time migration algorithm based on VTI media to improve the running efficiency of the algorithm.The main work contents include:(1)By analyzing the dependence between the data of the reverse time migration algorithm based on VTI media,a multi-level parallel optimization strategy(grid point,domain division,shot interval)is proposed to mine the parallelism of the algorithm and optimize the problem of the long time consuming of the large-scale reverse time migration algorithm.Through three groups of experiments,the optimization effect of this strategy is verified.The experimental results show that the multi-level parallel optimization strategy has obvious acceleration effect.A series of large-scale experiments are carried out to verify that the multi-level parallel optimization strategy eliminates the limitation of single node memory on the scale of the algorithm.(2)A parallel read and write optimization strategy is proposed.On the basis of the multi-level parallel optimization strategy,the parallel efficiency decreases rapidly with the increase of the number of processes due to the high ratio of read and write time.In order to solve this problem,based on the multi-level parallel optimization,this paper changes the storage structure of files according to the independent process of reading and writing between program blocks,and uses MPI technology to realize the parallel reading and writing between data blocks,so as to improve the efficiency of the algorithm.In addition,three groups of experiments are carried out on Dawning supercomputer platform,and the experimental data show that the parallel read and write strategy in this paper has a good acceleration effect,and the acceleration effect has a good stability.(3)On the Dawning supercomputer platform,this paper carried out comprehensive experiments on the optimized VTI media reverse time migration algorithm from three aspects.Firstly,the number of processes is increased while the total computing scale remains unchanged,and the acceleration effect of the optimized algorithm is studied.The experimental results show that the optimization strategy proposed in this paper has obvious acceleration effect.Secondly,the algorithm scalability is studied by increasing the number of processes under the condition that the node load remains unchanged.The experimental results show that the optimization strategy proposed in this paper has good scalability,and the greater the node load,the better the effect.Finally,a series of large-scale experiments are carried out to verify that the optimization strategy in this paper eliminates the limitation of single node memory on the algorithm scale.